Research map / Optimization / Optimizers
Momentum Centering and Asynchronous Update for Adaptive Gradient Methods
Juntang Zhuang, Yifan Ding, Tommy Ming Tang, Nicha C Dvornek, sekhar tatikonda, James s Duncan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use asynchronous centering instead of synchronous centering | AdaBelief-style adaptive optimization · ImageNet | Top-1 accuracy: 70.08$\ddag$ -> \textbf{70.46} | Improved |
| Use asynchronous centering instead of synchronous centering | Transformer · DE-EN | BLEU: 35.17$\pm$0.015 -> \textbf{35.35$\pm$0.012} | Improved |
| Use asynchronous centering instead of synchronous centering | Transformer · JA-EN | BLEU: \textbf{34.38$\pm$0.009} -> 33.70$\pm$0.021 | Hurt |
| Use asynchronous centering instead of synchronous centering | GAN training · DCGAN | FID: 47.25$\pm$0.79 -> \textbf{43.43$\pm$4.38} | Improved |
| Use asynchronous centering instead of synchronous centering | GAN training · RLGAN | FID: \textbf{36.58$\pm$0.12} -> 37.15$\pm$0.13 | Hurt |
| Use asynchronous centering instead of synchronous centering | GAN training · SNGAN | FID: 12.70$\pm$0.17 -> \textbf{12.44$\pm$0.02} | Improved |
| Use asynchronous centering instead of synchronous centering | GAN training · SAGAN | FID: 13.98$\pm$0.02 -> \textbf{13.54$\pm$0.15} | Improved |
| Use asynchronous updates rather than synchronous updates | Adaptive optimizers on the Reddi et al. periodic convex example · Online convex optimization | convergence | Improved |
| Add centered second momentum to asynchronous updates (ablation) | AdaShift · Sparse-gradient periodic convex problem | convergence | Improved |
| Use centered rather than uncentered second momentum (ablation) | Asynchronous adaptive optimization · Stochastic non-convex optimization | upper bound on regret | Improved |
Closest papers by meaning
- Towards Understanding Acceleration Tradeoff between Momentum and Asynchrony in Nonconvex Stochastic Optimization (NeurIPS 2018 · similarity 0.87)
- Improved Asynchronous Parallel Optimization Analysis for Stochastic Incremental Methods (JMLR 2018 · similarity 0.87)
- Ordered Momentum for Asynchronous SGD (NeurIPS 2024 · similarity 0.86)
- Faster Stochastic Optimization with Arbitrary Delays via Adaptive Asynchronous Mini-Batching (ICML 2025 · similarity 0.85)
- Rescaled Asynchronous SGD: Optimal Distributed Optimization under Data and System Heterogeneity (NeurIPS 2026 · similarity 0.85)
- Stability and Generalization of Asynchronous SGD: Sharper Bounds Beyond Lipschitz and Smoothness (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
